Discrete Mathematics and Theoretical Computer Science DMTCS vol. 19:3, 2017, #10 Binary Codes and Period-2 Orbits of Sequential Dynamical Systems arxiv:1509.03907v5 [math.co] 30 Sep 2017 1 Princeton University 2 University of Florida Colin Defant 1,2 received 19 th Jan. 2017, revised 30 th May 2017, 16 th Sep. 2017, accepted 20 th Sep. 2017. Let [K n, f, π] be the (global) SDS map of a sequential dynamical system (SDS) defined over the complete graph K n using the update order π S n in which all vertex functions are equal to the same function f : F n 2 F n 2. Let η n denote the maximum number of periodic orbits of period 2 that an SDS map of the form [K n, f, π] can have. We show that η n is equal to the maximum number of codewords in a binary code of length n 1 with minimum distance at least 3. This result is significant because it represents the first interpretation of this fascinating coding-theoretic sequence other than its original definition. Keywords: Sequential dynamical system; periodic orbit; complete graph; binary code. 1 Introduction Suppose we wish to model a finite system in which objects have various states and update their states in discrete time steps. Moreover, assume that the state to which an object updates depends only on the current state of that object along with the states of other nearby or connected objects. We can capture such a system s behavior with a graph dynamical system. A graph dynamical system contains a graph representing the connections between objects, a set of states that the objects can adopt, a collection of functions that model how each individual object updates its state in reaction to its neighbor s states, and a rule determining the scheme by which the objects update their states. In a series of papers published between 1999 and 2001, Barrett, Mortveit, and Reidys introduced the notion of a sequential dynamical system (SDS), a graph dynamical system in which vertices update their states sequentially [9, 6, 7]. Subsequently, several researchers have worked to develop a general theory of SDS (see, for example, [3, 4, 5, 10, 11, 12, 13, 19]). The article [10] is interesting because it shows how SDS, originally proposed as models of computer simulation, are now being studied in relation with Hecke-Kiselman monoids in algebraic combinatorics. We draw most of our terminology and background information concerning SDS from [16], a valuable reference for anyone interested in exploring this field. In the theory of SDS, the primary focus of many research articles is to count or otherwise characterize periodic orbits in the phase spaces of sequential dynamical systems [1, 2, 4, 8, 21, 22, 23]. For example, This work was supported by National Science Foundation grant DMS-1358884. ISSN 1365 8050 c 2017 by the author(s) Distributed under a Creative Commons Attribution 4.0 International License
2 Colin Defant the recent paper [1] studies which periodic orbits can coexist in certain SDS and when certain SDS must necessarily have unique fixed points. In particular, that article shows that analogues of Sharkovsky s theorem from continuous dynamics completely fail to hold for many families of SDS. It is common to analyze the dynamics of sequential dynamical systems defined using classical Boolean functions such and OR, AND, NOR, and NAND (see, for example, [4, 8]). The article [23] focuses more generally on SDS defined using so-called bi-threshold functions. By contrast, we will consider SDS defined using a completely arbitrary update function f. As we describe later, this function will be the vertex function for every vertex in the graph (this is possible because we will only consider base graphs that are complete). We now proceed to clarify some of these remarks by establishing some notation and definitions. If v is a vertex of a graph Y, we let d(v) denote the degree of v. We often work in the finite field F 2 = {0, 1}. In doing so, we let x = 1 + x for any x F 2. For any vector x = (x 1, x 2,..., x k ) F k 2, let inv( x) = (x 1, x 2,..., x k ). Furthermore, id will denote the identity permutation 123 n (the length n of the permutation id will always be clear from context). An SDS is built from the following parts: An undirected simple graph Y with vertices v 1, v 2,..., v n. A set of states A. We will typically use the set of states A = F 2 = {0, 1}. A collection of vertex functions {f vi } n i=1. Each vertex v i of Y is endowed with its own vertex function f vi : A d(vi)+1 A. A permutation π S n. The permutation π is known as the update order. Let q(v) denote the state of a vertex v. Suppose a vertex v i has neighbors v j1, v j2,..., v jd(vi ), where j 1 < j 2 < < j s < i < j s+1 < j s+2 < < j d(vi). We let X(v i ) = (q(v j1 ), q(v j2 ),..., q(v js ), q(v i ), q(v js+1 ), q(v js+2 ),..., q(v jd(vi ) )). For example, if the vertex v 3 has neighbors v 1, v 4, and v 6, we let X(v 3 ) = (q(v 1 ), q(v 3 ), q(v 4 ), q(v 6 )). The vector (q(v 1 ), q(v 2 ),..., q(v n )), which lists all of the states of the vertices of Y in the order corresponding to the order of the vertex indices, is known as the system state of the SDS. Note that if Y is a complete graph and v i is any vertex of Y, then X(v i ) is equal to the system state of the SDS. From each vertex function f vi, define the local update function L vi : A n A n by L vi (x 1, x 2,..., x n ) = (x 1, x 2,..., x i 1, f vi (X(v i )), x i+1,..., x n ). Combining these local update functions with the update order π = π(1)π(2) π(n) (we have written the permutation π as a word), we obtain the SDS map F : A n A n given by F = L vπ(n) L vπ(n 1) L vπ(1). We will find it useful to introduce an intermediate SDS map G i : A n A n for each i {1, 2,..., n}, which we define by G i = L vπ(i) L vπ(i 1) L vπ(1). Thus, F = G n. We use the convention that G 0 denotes the identity map from A n to A n. The vector G i ( x) represents the system state of the SDS obtained by starting with a system state x and updating only the
Binary Codes and Period-2 Orbits 3 v 2 A 0 B C 1 D 1 E v 1 0 0 1 v 3 v 4 v 2 v 1 0 0 1 v 3 v 4 v 2 v 1 0 0 1 v 3 v 4 v 2 v 1 0 0 1 v 3 v 4 1 v 2 v 1 1 1 v 3 v 4 Fig. 1: A system update of the SDS of Example 1.1. Block A shows the inital state of the SDS. Blocks B, C, and D show the intermediate steps of the system update. Block E shows the system state obtained after completing the system update. first i vertices in the update order π. Once the system updates all n vertices (known as a system update), the new system state is F ( x). Given any SDS on a graph Y with vertex functions {f vi } n i=1 and update order π, we denote its SDS map F by the triple [Y, {f vi } n i=1, π]. If all of the vertex functions f v i are equal to the same function f, we will simply write [Y, f, π] for the corresponding SDS map (this situation can only occur if the base graph is regular). Example 1.1. Consider the graph Y shown in block A of Figure 1. We define an SDS over Y using the update order π = 2413 and the vertex functions f vi given by and f v1 (x 1, x 2, x 3, x 4 ) = x 1 x 3 + x 2 + x 4, f v2 (x 1, x 2 ) = x 1 x 2 + 1, f v3 (x 1, x 2, x 3 ) = x 1 + x 2 + x 3, f v4 (x 1, x 2, x 3 ) = x 1 x 2 + x 3. The initial system state of this SDS is (0, 0, 0, 1), as shown by the blue labels in block A of Figure 1. Because π(1) = 2, we first update the vertex v 2 using the vertex function f v2. We have f v2 (X(v 2 )) = f v2 (q(v 1 ), q(v 2 )) = f v2 (0, 0) = 1, so the vertex v 2 updates to the new state 1. This intermediate update is shown in block B of Figure 1. Another way to understand the transition from block A to block B in the figure is to see that we have changed the system state of the SDS by applying the local update function L v2. Indeed, L v2 (0, 0, 0, 1) = (0, 1, 0, 1). In a similar fashion, we next update vertices v 4, v 1, and v 3. Letting F = [Y, {f vi } 4 i=1, π], we find that F (0, 0, 0, 1) = (0, 1, 1, 1), as shown in block E of the figure. In other words, through a sequence of local updates, the system update transformed the system state (0, 0, 0, 1) into the new system state (0, 1, 1, 1). The SDS map F tells us how the states of the vertices of the graph Y change when we update the graph in a sequential manner. A useful tool for visualizing how F acts on the system s states is the phase space of the SDS. The phase space, denoted Γ(F ), is the directed graph with vertex set V (Γ(F )) = A n and edge set E(Γ(F )) = {( x, y) A n A n : y = F ( x)}.
4 Colin Defant 0110 0010 1010 1011 1100 1111 1101 0000 1000 1001 0100 1110 0001 0111 0101 0011 Fig. 2: The phase space of the SDS described in Example 1.1. Note that we have omitted parentheses and commas from the vectors in order to improve the aesthetics of the image. For example, the vector (1, 0, 1, 1) is written as 1011. In other words, we draw a directed edge from x to F ( x) for each x A n. As an example, the phase space of the SDS given in Example 1.1 is shown in Figure 2. Notice that the phase space shown in Figure 2 has a single 2-cycle (formed from the vertices 0111 and 0101). In general, a phase space of an SDS can have many cycles of various lengths. This leads us to some interesting questions. For example, when is it possible to have a phase space composed entirely of 2-cycles? Can we find an upper bound on the number of 2-cycles that can appear in the phase spaces of certain SDS? If we can show, for instance, that certain SDS defined over a graph with n vertices cannot have phase spaces consisting entirely of 2-cycles, then it will follow that the very natural function inv: F n 2 F n 2 cannot be the SDS map of any of those SDS. In this paper, we study the number of 2-cycles that can appear in the phase spaces of SDS defined over a complete graph K n in which all vertex functions are the same. More formally, we give the following definition. Definition 1.1. For a positive integer n, let (F 2 ) Fn 2 be the set of all functions g : F n 2 F 2. For each g (F 2 ) Fn 2 and π Sn, let η(g, π) denote the number of 2-cycles in the phase space Γ([K n, g, π]). Define η n = max g (F 2) Fn 2 η(g, id). Remark 1.1. Our decision to use the identity update order id in the definition of η n in Definition 1.1 stems from a desire for convenience, but we lose no generality in making such a decision because we are working over complete graphs. In other words, η n = max η(g, id) = max η(g, π). g (F 2) Fn 2 g (F 2) Fn 2 π S n In the next section, we reformulate the problem of determining η n in terms of finding the clique number of a certain graph. We then show that η n+1 = A(n, 3), where A(n, 3) denotes the maximum number of
Binary Codes and Period-2 Orbits 5 codewords in a binary code of length n with minimum distance at least 3. The sequence A(n, 3), which is sequence A005864 in Sloane s Online Encyclopedia of Integer Sequences [17], has been a fascinating and mysterious subject of inquiry in coding theory [14, 15, 18, 20]. Our result is noteworthy because, to the best of our knowledge, it provides the only known interpretation of this sequence other than its original definition. 2 Searching for η n If X is a set and F : X X is a function, we say an element x of X is a periodic point of period 2 of F if F 2 ( x) = x and F ( x) x. Typically, we use the field F 2 as our set of states. However, in the following lemma, we may use any set of states A so long as A 2. Lemma 2.1. Let n 2 be an integer. Let f : A n A be a function, and let π = π(1)π(2) π(n) S n be a permutation. Suppose x = (x 1, x 2,..., x n ) is a periodic point of period 2 of the SDS map F = [K n, f, π]. Write F ( x) = z = (z 1, z 2,..., z n ). For each k {1, 2,..., n}, we have x k z k. In particular, if A = F 2, then F ( x) = inv( x). Proof: For any j {1, 2,..., n}, recall that G j 1 ( x) is the system state that results from starting with the initial system state x and then updating the vertices v π(1), v π(2),..., v π(j 1) (in this order). When we update the state of the vertex v π(j), the state to which v π(j) updates is given by f(g j 1 ( x)). On the other hand, the state to which v π(j) updates must be z π(j), the state of v π(j) in the system state z. It follows that f(g j 1 ( x)) = z π(j). Note that f(g n ( x)) = f(f ( x)) = f( z) = f(g 0 ( z)) = x π(1). These same arguments show that f(g j 1 ( z)) = x π(j) for any j {1, 2,..., n} and that f(g n ( z)) = z π(1). If i {1, 2,..., n 1}, then the preceding paragraph tells us that f(g i 1 ( x)) = z π(i), f(g i 1 ( z)) = x π(i), We also have f(g i ( x)) = z π(i+1), f(g i ( z)) = x π(i+1). (1) f(g n 1 ( x)) = z π(n), f(g n 1 ( z)) = x π(n), f(g n ( x)) = x 1, f(g n ( z)) = z 1. (2) Notice that if x π(i) = z π(i), then updating the vertex v π(i) doesn t change anything; that is, G i 1 ( x) = G i ( x) and G i 1 ( z) = G i ( z). With the help of the equations in (1) and (2), this shows that if x π(i) = z π(i) for some i {1, 2,..., n}, then x π(i+1 (mod n)) = z π(i+1 (mod n)). As a consequence, we see that if x π(i) = z π(i) for some i {1, 2,..., n}, then x = z = F ( x). However, we are assuming that x is a periodic point of F of period 2, so F ( x) x by definition. Thus, x π(i) z π(i) for all i {1, 2,..., n}, which proves the lemma. Let w = w 1 w 2... w k be a finite word over the alphabet F 2. We say that a vector (x 1, x 2,..., x n ) F n 2 contains the subsequence w if there exist i 1, i 2,..., i k {1, 2,..., n} such that i 1 < i 2 < < i k
6 Colin Defant and x ij = w j for all j {1, 2,..., k} (this is sometimes expressed by saying that w is a scattered subword of (x 1, x 2,..., x n )). For example, the vector (x 1, x 2, x 3, x 4, x 5 ) = (1, 0, 1, 1, 0) contains the subsequence 100 because x 1 = 1, x 2 = 0, and x 5 = 0. However, the vector (1, 0, 1, 1, 0) does not contain the subsequence 001. This leads us to the following definition. Definition 2.1. Let w be a finite word over the alphabet F 2. Define D n (w) to be the set of vectors x F n 2 such that x contains the subsequence w. Definition 2.2. Let n 2 be an integer. Let F n 2 = {(x 1, x 2,..., x n ) F n 2 : x 1 = 0}. Define Ĥn to be the undirected simple graph with vertex set and edge set V (Ĥn) = F n 2 E(Ĥn) = {{ x, y} F n 2 : x + y D n (101)}. The following two lemmas link the graphs Ĥn to our study of 2-cycles in the phase spaces of SDS defined over complete graphs. Lemma 2.2. Let n 2 be an integer, and let C = { x 1, x 2,..., x k } be a clique of order k of Ĥn. There exists a map f : F n 2 F 2 such that each x i C is in a 2-cycle of Γ([K n, f, id]). Moreover, no two distinct x i, x j are contained in the same 2-cycle of Γ([K n, f, id]). Proof: For each i {1, 2,..., k}, let x i = (a i1, a i2,..., a in ), where a i1 = 0 by the definition of Ĥn. Define the map f : F n 2 F 2 as follows: If there are some i, l such that then let f( α) = a i(l+1). If there are some i, l such that then let f( α) = a i(l+1). Otherwise, let f( α) = 0. α = (a i1, a i2,..., a il, a i(l+1), a i(l+2),..., a in ), α = (a i1, a i2,..., a il, a i(l+1), a i(l+2),..., a in ), We first need to show that f is well-defined. To do so, we show that for any i, j {1, 2,..., k} and any l, m {0, 1,..., n 1} with i j or l m, we have (a i1,..., a il, a i(l+1),..., a in ) (a j1,..., a jm, a j(m+1),..., a jn ), (3) (a i1,..., a il, a i(l+1),..., a in ) (a j1,..., a jm, a j(m+1),..., a jn ), (4)
Binary Codes and Period-2 Orbits 7 and (a il,..., a il, a i(l+1),..., a in ) (a j1,..., a jm, a j(m+1),..., a jn ). (5) This will show that we have not accidentally defined f( α) = 0 and f( α) = 1 for the same vector α. It is easy to see that the lack of equality must hold in (3), (4), and (5) if i = j and l m. Thus, we may assume i j. If equality holds in (3) or (5), then x i + x j = (0, 0,..., 0, 1,..., 1, 0,..., 0) D n (101), which is a contradiction. Similarly, if equality holds in (4), then it follows from the fact that a i1 = a j1 = 0 that x i +x j = (0, 0,..., 0, 1, 1,..., 1) D n (101). This is again a contradiction. Thus, f is well-defined. Let F = [K n, f, id]. We now show that for each i {1, 2,..., k}, the vector x i is in a 2-cycle of Γ(F ). Choose some i {1, 2,..., k}. It follows from the definition of f that for each l {1, 2,..., n} and G l ( x i ) = (a i1,..., a il, a i(l+1),..., a in ) G l (inv( x i )) = G l (a i1, a i2,..., a in ) = (a i1,..., a il, a i(l+1),..., a in ). In particular, F ( x i ) = inv( x i ) x i and F 2 ( x i ) = F (inv( x i )) = x i. In other words, x i is in a 2-cycle of Γ(F ). Finally, choose some distinct i, j {1, 2,..., k}. It is easy to see that the vectors x i and x j are in distinct 2-cycles of Γ(F ). Since x i x j, the only way the vectors x i and x j could be in the same 2-cycle of Γ(F ) is if F ( x i ) = x j. We have just shown that F ( x i ) = inv( x i ), so x i could only be in the same 2- cycle as x j if x j = inv( x i ). However, this is impossible because x i and x j have the same first coordinate (namely, 0). Lemma 2.3. Let n 2 be an integer, and let x and y be distinct nonadjacent vertices of Ĥ n. Let f : F n 2 F 2 be a function. If x is in a 2-cycle of the phase space Γ([K n, f, id]), then y is not in a 2-cycle of Γ([K n, f, id]). Proof: Suppose, by way of contradiction, that x and y are both in 2-cycles of Γ([K n, f, id]). Let F = [K n, f, id]. Let x = (a 1, a 2,..., a n ) and y = (b 1, b 2,..., b n ). Because x y, we may let r be the smallest element of {1,..., n} such that a r b r. Similarly, we may let s be the largest element of {1,..., n} such that a s b s. Note that r 2 because a 1 = b 1 = 0. Because x and y are not adjacent in Ĥ n, the vector x + y = (a 1 + b 1,..., a n + b n ) does not contain the subsequence 101. This implies that { 1, if r i s; a i + b i = 0, otherwise. In other words, y = (a 1,..., a r 1, a r,..., a s, a s+1,..., a n ). For the sake of convenience, let a n+1 = a 1. Because we are assuming that each of the vectors x and y is in a 2-cycle of Γ(F ), we know from Lemma 2.1 that F ( x) = inv( x) and F ( y) = inv( y). Hence, for each l {1,..., n}, G l ( x) = (a 1,..., a l, a l+1,..., a n ) and G l ( y) = (b 1,..., b l, b l+1,..., b n ).
8 Colin Defant It follows that but also f(a 1,..., a s, a s+1..., a n ) = f(g s ( x)) = a s+1, f(a 1,..., a s, a s+1..., a n ) = f(b 1,..., b r 1, b r,..., b n ) = f(g r 1 ( y)) = b r = a r. Therefore, a r = a s+1. Similarly, we have and f(a 1,..., a r 1, a r..., a n ) = f(g r 1 ( x)) = a r f(a 1,..., a r 1, a r..., a n ) = f(b 1,..., b s, b s+1,..., b n ) = f(g s ( y)) = b s+1 = a s+1. This implies that a r = a s+1 = a r, which is a contradiction. We are now in a position to prove one of our crucial theorems. Let ω(g) denote the clique number of a graph G. That is, G contains a clique of order ω(g) but does not contains a clique of order ω(g) + 1. Theorem 2.1. For any integer n 2, we have η n = ω(ĥn). Proof: Choose an integer n 2, and let k = ω(ĥn). Let C = { x 1, x 2,..., x k } be a clique of order k of Ĥn. By Lemma 2.2, there exists a map f : F n 2 F 2 such that any distinct vectors x i, x j C are in distinct 2-cycles of Γ([K n, f, id]). In particular, Γ([K n, f, id]) contains at least k 2-cycles. In the notation of Definition 1.1, f (F 2 ) Fn 2, and η(f, id) k. Thus, η n = max η(g, id) η(f, id) k. g (F 2) Fn 2 We now show that η n k. By Definition 1.1, there exists a function g (F 2 ) Fn 2 such that η(g, id) = η n. In other words, there are η n 2-cycles in the phase space Γ([K n, g, id]). It follows from Lemma 2.1 that each 2-cycle of Γ([K n, g, id]) contains exactly one vector whose first coordinate is 0. In the notation of Definition 2.2, each of the η n 2-cycles of Γ([K n, g, id]) contains exactly one vector that is a vertex of Ĥ n. Let x 1, x 2,..., x ηn be these vertices of Ĥn. Choose some distinct i, j {1, 2,..., η n }. Because each of the vectors x i, x j is in a 2-cycle of Γ([K n, g, id]), it follows from Lemma 2.3 that x i and x j must be adjacent in Ĥn. Because i and j were arbitrary, this shows that { x 1, x 2,..., x ηn } is a clique of Ĥn. Consequently, η n k. Given an integer n 2, we may define a map θ n : F n 2 F n 1 2 by θ n (x 1, x 2,..., x n ) = (x 2, x 3,..., x n ). Because the first coordinate of each vector in F n 2 is 0, it should be clear that θ n is a vector space isomorphism. Furthermore, if x, y F n 2, then x + y D n (101) if and only if θ n ( x) + θ n ( y) D n 1 (101). This motivates the following definition.
Binary Codes and Period-2 Orbits 9 Definition 2.3. Given an integer m 2, define H m to be the undirected simple graph with vertex set V (H m ) = F m 2 and edge set E(H m ) = {{ x, y} F m 2 : x + y D m (101)}. It follows from the preceding paragraph that θ n+1 is a graph isomorphism from the graph Ĥn+1 to the graph H n. Therefore, η n+1 = ω(ĥn+1) = ω(h n ). In the following section, we relate the graphs H n to error-correcting binary codes. 3 Binary Codes In coding theory, a binary code of length n is a subset of F n 2. The elements of a code are known as codewords. The Hamming distance between two codewords x = (x 1, x 2,..., x n ) and y = (y 1, y 2,..., y n ), which we shall denote by δ( x, y), is simply the number of positions in which the vectors x and y have different coordinates. That is, δ( x, y) = {i {1, 2,..., n}: x i y i }. If C is a nonempty binary code of length n, then the minimum distance of C, denoted (C), is the quantity (C) = min δ( x, y). x, y C x y We make the convention that (C) = if the code C consists of a single codeword. Of particular interest in coding theory are binary codes with minimum distance at least 3. Such codes are known as one-bit error-correcting codes because any error formed by flipping a single bit (that is, changing a single coordinate) in a codeword can be detected and corrected (this is not the case for binary codes with minimum distance less than 3). Let A(n, 3) denote the maximum number of codewords that can appear in a binary code of length n that has minimum distance at least 3. An important problem in coding theory is the determination of the values of A(n, 3). It is known that A(n, 3) = 2 n log 2 (n+1) if n + 1 is a power of 2, but most values of A(n, 3) are not known. Given an integer m 2, define J m to be the undirected simple graph with vertex set and edge set V (J m ) = F m 2 E(J m ) = {{ x, y} F m 2 : x + y D m (111)}. Observe that two vectors x, y F m 2 are adjacent in J m if and only if the Hamming distance δ( x, y) between x and y is at least 3. Therefore, A(n, 3) is equal to the clique number ω(j n ) of the graph J n. From this, we obtain the following theorem. Theorem 3.1. For any positive integer n, η n+1 = A(n, 3).
10 Colin Defant Proof: We saw at the end of the preceding section that η n+1 = ω(h n ). Therefore, in light of the preceding paragraph, we see that it suffices to show that ω(h n ) = ω(j n ). Consider the linear transformation T : F n 2 F n 2 given by T (x 1, x 2,..., x n ) = (x 1, x 1 + x 2, x 1 + x 2 + x 3,..., x 1 + x 2 + + x n ). The linear tranformation T is an endomorphism with a trivial kernel, so it is an isomorphism. Furthermore, a vector z F n 2 contains the subsequence 111 if and only if T ( z) contains the subsequence 101. This shows that T is in fact a graph isomorphism from J n to H n, so ω(h n ) = ω(j n ). As mentioned in the introduction, Theorem 3.1 provides the first known interpretation of the numbers A(n, 3) outside of coding theory. 4 Concluding Remarks We have defined η n to be the maximum number of 2-cycles that can appear in a phase space of the form Γ([K n, f, id]). One could easily generalize these numbers by defining η n (m) to be the maximum number of m-cycles that can appear in such a phase space. One may show that η n (1) = 2 for all positive integers n (the only possible fixed points of an SDS map [K n, f, id] are the all-0 s and all-1 s vectors of length n). Is it possible to obtain general bounds for the numbers η n (m)? Could we perhaps relate the numbers η n (m) to codes as we have done for the numbers η n (2)? There are, of course, many other natural ways to generalize the problems considered here. One might wish to replace complete graphs with graphs that are, in some sense, almost complete (such as complements of cycle graphs). We could also choose to ask similar questions about SDS maps of the form [K n, f, id] defined using a set of states A with A 3. 5 Acknowledgments The author would like to thank Padraic Bartlett for introducing the author to sequential dynamical systems and giving excellent suggestions for the improvement of this paper. The author would also like to thank the anonymous DMTCS referees for several helpful suggestions. References [1] J. A. Aledo, L. G. Diaz, S. Martinez, and J. C. Valverde. On periods and equilibria of computational sequential systems. Information Sciences, Volumes 409 410 (2017), 27 34. [2] C. L. Barrett, W. Y. C. Chen, and M. J. Zheng. Discrete dynamical systems on graphs and Boolean functions. Math. Comput. Simul., 66 (2004) 487 497. [3] C. L. Barrett, H. B. Hunt III, M. V. Marathe, S. S. Ravi, D. J. Rosenkrantz, and R. E. Stearns. Reachability problems for sequential dynamical systems with threshold functions. Theoret. Comput. Sci., 295 (2003), 41 64. [4] C. L. Barrett, H. B. Hunt III, M. V. Marathe, S. S. Ravi, D. J. Rosenkrantz, R. E. Stearns, and P. T. Tosic. Gardens of Eden and fixed points in sequential dynamical systems. DM CCG, (2001), 95 110.
Binary Codes and Period-2 Orbits 11 [5] C. L. Barrett, H. B. Hunt III, M. V. Marathe, S. S. Ravi, D. J. Rosenkrantz, and R. E. Stearns. On some special classes of sequential dynamical systems. Ann. Comb., 7 (2003), 381 408. [6] C. L. Barrett, H. S. Mortveit, and C. M. Reidys. Elements of a theory of computer simulation II: sequential dynamical systems. Appl. Math. Comput., 107 (2000), 121 136. [7] C. L. Barrett, H. S. Mortveit, and C. M. Reidys. Elements of a theory of computer simulation III: equivalence of SDS. Appl. Math. Comput., 122 (2001), 325 340. [8] C. L. Barrett, H. S. Mortveit, and C. M. Reidys. Elements of a theory of computer simulation IV: sequential dynamical systems: fixed points, invertibility and equivalence. Appl. Math. Comput., 134 (2003), 153 171. [9] C. L. Barrett and C. M. Reidys. Elements of a theory of computer simulation I: Sequential CA over random graphs. Appl. Math. Comput., 98 (1999), 241 259. [10] E. Collina and A. D Andrea. A graph-dynamical interpretation of Kiselman s semigroups. J. Algebraic Combin., 41 (2015), 1115 1132. [11] L. D. Garcia, A. S. Jarrah, and R. Laubenbacher. Sequential dynamical systems over words. Appl. Math. Comput., 174 (2006), 500 510. [12] M. Macauley, J. McCammond, and H. S. Mortveit. Dynamics groups of asynchronous cellular automata. J. Algebraic Combin., 33 (2011), 11 35. [13] M. Macauley and H. S. Mortveit. Cycle Equivalence of Graph Dynamical Systems. Nonlinearity, 22 (2009), 421-436. [14] F. J. MacWilliams and N. J. A. Sloane, The theory of error-correcting codes. Elsevier-North Holland, 1978, p. 674. [15] M. Milshtein, A new binary code of length 16 and minimum distance 3, Information Processing Letters, 115.12 (2015), 975 976. [16] H. S. Mortveit and C. M. Reidys. An introduction to sequential dynamical systems. Springer Science & Business Media, 2007. [17] The On-Line Encyclopedia of Integer Sequences, published electronically at https://oeis.org. [18] P. R. J. Ostergard, T. Baicheva, and E. Kolev, Optimal binary one-error-correcting codes of length 10 have 72 codewords, IEEE Trans. Inform. Theory, 45 (1999), 1229 1231. [19] C. M. Reidys. On acyclic orientations and sequential dynamical systems. Adv. Appl. Math., 27 (2001), 790 804. [20] A. M. Romanov. New binary codes of minimal distance 3, Problemy Peredachi Informatsii, 19 (1983), 101 102. [21] P. T. Tosic and G.U. Agha. On computational complexity of counting fixed points in symmetric Boolean graph automata. Lect. Notes Comput. Sci. 3699 (2005) 191 205.
12 Colin Defant [22] A. Veliz-Cuba and R. Laubenbacher. On computation of fixed points in Boolean networks, J. Appl. Math. Comput. 39 (2012) 145 153. [23] S. Wu, A. Adig, and H.S. Mortveit. Limit cycle structure for dynamic bi-threshold systems. Theor. Comput. Sci. 559 (2014) 34 41.